AI Budget Raids Hit Mainframe Refresh Cycles in 2026

What IBM's numbers actually show

IBM published preliminary Q2 figures ahead of its full earnings release, alongside an unusually candid letter from CEO Arvind Krishna. Infrastructure revenue fell 7 percent, a sharp reversal for a segment IBM had previously billed as enjoying the strongest mainframe launch in its history. Krishna's explanation wasn't waning appetite for Z-series hardware but a late-June scramble by clients to lock in servers, storage and memory before prices climbed further and supply tightened.

The knock-on effects went beyond hardware. Fewer mainframe deals meant weaker sales of the transaction-processing software that typically rides alongside a Z refresh, dragging IBM's software numbers down too. Krishna also flagged "rapidly evolving, industry-wide cybersecurity concerns" as a distraction during the quarter, though he gave no detail on what these were.

Why UK buyers are quietly raiding legacy budgets

This isn't an isolated US phenomenon. UK enterprises likely face similar global supply constraints — inferred to include GPUs, high-bandwidth memory and enterprise SSDs, the component categories generally cited as supply-constrained — and are likely making a comparable calculation: buy now, at today's price, or risk paying considerably more later, or not being able to buy at all. Internal analysis from our AI Server Cost Index 2026 research — an in-house dataset that has not been independently verified externally — suggests similar pricing pressure is driving this behaviour across UK procurement teams.

The logic is understandable at a line-item level. A planned mainframe refresh has a predictable timeline and a known vendor relationship. AI infrastructure, by contrast, is sitting in a market where component shortages and price volatility make delay genuinely costly. Faced with that asymmetry, finance teams are pulling capex forward into whatever hardware feels most exposed to near-term inflation — which right now is servers, memory and storage rather than mainframes.

The refresh cycle risk nobody's budgeting for

The danger for UK buyers is that this reallocation isn't a strategic decision — it's a reaction. Deferring a planned mainframe or legacy server refresh to fund AI stockpiling doesn't remove the underlying need; it just pushes it to a later, possibly more expensive, quarter. IBM's own admission that "numerous large deals failed to close on the timelines" it expected is a warning sign for any buyer assuming vendors can simply absorb a delayed order without consequence.

Teams reworking their IT finance calculator models this year should treat legacy refresh slippage as a line item in its own right, not an afterthought. A deferred mainframe cycle can quietly extend support costs, licensing exposure and failure risk on ageing kit well beyond what the AI budget reallocation was meant to save.

TCO planning has to catch up with the panic buying

Total cost of ownership models built a year ago assumed relatively stable hardware pricing. That assumption no longer holds. Our own analysis, published separately, of why AI servers cost more and the broader server memory and SSD price surge — findings that have not been independently verified externally — points to a similar structural driver behind IBM's quarter: components shared between AI and general-purpose enterprise servers are being bid up simultaneously, and buyers who wait are absorbing the difference. See our related coverage on why AI servers cost more and the broader server memory and SSD price surge.

For UK procurement teams, the practical question isn't whether to buy — it's whether the current refresh plan still reflects real supply and pricing conditions. Running the numbers through an AI GPU Calculator alongside a legacy TCO model gives a clearer picture of where budget is genuinely best spent this quarter versus where it's simply being moved out of fear.

Not every part of IBM's business suffered

It's worth noting the reallocation wasn't uniformly bad news for IBM. Red Hat revenue grew 11 percent, recent acquisitions including HashiCorp and Confluent performed strongly, and Distributed Infrastructure — covering Power servers and storage — posted record reported growth of 37 percent. That pattern mirrors what UK buyers are doing: money isn't disappearing from enterprise IT, it's migrating toward the hardware and software categories seen as most exposed to near-term scarcity.

What UK buyers should do before the next quarter

If capex is being reprioritised anyway, it pays to do it deliberately rather than reactively. That means getting a firm read on current lead times and pricing before committing, comparing configurations across Dell, HPE and Lenovo platforms, and stress-testing whether it's smarter to buy servers now or wait given your specific refresh timeline. Buyers weighing a formal purchase should also get a like-for-like enterprise server quote before assuming today's list price will hold.